This repository contains the Multi-SWE-bench dataset, introduced in Multi-SWE-bench: A Multilingual Benchmark for Issue Resolving, to address the lack of multilingual benchmarks for evaluating LLMs in real-world code issue resolution.
Unlike existing Python-centric benchmarks (e.g., SWE-bench), this framework spans 7 languages (Java, TypeScript, JavaScript, Go, Rust, C, and C++) with 1,632 high-quality instances,
curated from 2,456 candidates by 68 expert annotators… See the full description on the dataset page:
https://huggingface.co/datasets/ByteDance-Seed/Multi-SWE-bench-flash.
